Supersonic aircraft aerodynamic performance prediction based on machine learning approach

This study proposes a novel approach that integrates machine learning algorithms with computational fluid dynamics (CFD) simulations to efficiently predict the aerodynamic performance of supersonic aircraft under cruising flight conditions. Three intelligent models were evaluated based on their principles, advantages, and limitations and then applied to predict the aerodynamic performance of a reference supersonic passenger aircraft. Bayesian optimization algorithms were employed on the selected models to enhance prediction accuracy. The performance metrics show that the proposed models can accurately predict supersonic aerodynamics under various flight conditions. In particular, the XGBoost model has the highest predictive accuracy for lift and drag coefficients, and the Bagging model is excellent at predicting the pitch moment coefficient. Machine learning-based prediction methods are more efficient than traditional CFD simulations, providing faster results without sacrificing accuracy. Moreover, Shapley Additive explanation was applied to the optimal model, which improves interpretability and offers deeper insights into the aerodynamic optimization of supersonic aircrafts.